粒子群径向基函数人工智能网络

发布时间：2018-03-16 00:03

本文选题：群智能　切入点：粒子群优化算法　出处：《东北石油大学》2017年硕士论文　论文类型：学位论文

【摘要】：本文主要研究了粒子群优化算法(PSO)和径向基函数神经网络(RBF)两方面内容。作为群智能演化算法的一种,粒子群凭借其原理简单易于实现,同时能够处理一些传统方法不能处理的问题,如不可导的节点传递函数或者梯度信息不存在之类的问题,广泛应用于神经网络训练、函数优化、模糊系统控制等诸多领域并取得了较好的效果。RBF神经网络是一种前馈神经网络,可以很好地克服传统神经网络学习过程的收敛过分依赖于初值和可能出现局部收敛的缺陷,具有较快的运算速度、较强的非线性映射能力和较好的预测效能,因而在储层参数预测中具有广阔的应用前景。本文将PSO算法与RBF神经网络有机结合,构建一个结构简单,稳定性强,鲁棒性高的人工智能神经网络,并结合实际地震资料,提取地震属性,预测储层参数。粒子群优化算法通过各粒子之间相互协作并共享信息对问题进行优化,其优势在于简单容易实现并且没有许多参数的调节。但是,标准的粒子群优化算法也存在早熟收敛、局部寻优能力较差等缺陷。针对以上不足,从更新公式,优化方式以及基于认知多样性变异三方面对算法进行改进。在更新公式中引用线性递减的惯性权重系数以及柯西分布和高斯分布的随机函数;采用逆序协同优化异步实现策略提高算法解决更高维度问题的能力;充分利用个体认知多样性,使个体最优以一定概率发生变异,带领群体逃离局部最优,收敛到全局。Benchmark function测试表明改进的粒子群算法优于其他优化算法。模型数据和实际地震数据反演精度较高、反演剖面横向连续性好、纵向分辨率较高、层间内部沉积、断层和分界面刻画清晰,证实了该算法具有很强的全局搜索能力和抗噪能力。本文首先对输入属性样本的预处理本文采用了起降维作用的k-l变换,然后模糊聚类分析提供隐层节点数、基中心和方差等网络参数,最后利用改进的粒子群算法优化出网络权值。标准测试集数据和砂体厚度预测的实际应用证实了该网络的可行性和实用价值,同时预测精度较高。
[Abstract]:Two aspects of particle swarm optimization (PSO) and radial basis function neural network (RBF) are studied in this paper. As a kind of swarm intelligence evolution algorithm, particle swarm optimization is simple and easy to implement. At the same time, it can deal with some problems that can not be dealt with by traditional methods, such as the non-differentiable node transfer function or the non-existence of gradient information. It is widely used in neural network training, function optimization, and so on. The RBF neural network is a kind of feedforward neural network, which can overcome the defects that the convergence of the traditional neural network learning process depends on the initial value and may appear local convergence. Because of its fast operation speed, strong nonlinear mapping ability and better prediction efficiency, it has a broad application prospect in reservoir parameter prediction. This paper combines PSO algorithm with RBF neural network to construct a simple structure. Artificial intelligence neural network with strong stability and high robustness is used to extract seismic attributes and predict reservoir parameters in combination with actual seismic data. Particle swarm optimization algorithm optimizes the problem by cooperating with each other and sharing information. The advantage of PSO is that it is simple and easy to realize and has no adjustment of many parameters. However, the standard PSO algorithm also has some defects such as premature convergence and poor local optimization ability. In the updating formula, the inertial weight coefficient of linear decrement and the random function of Cauchy distribution and Gao Si distribution are used in the updating formula. The asynchronous implementation strategy of inverse sequence cooperative optimization is adopted to improve the ability of the algorithm to solve higher dimensional problems, to make full use of individual cognitive diversity, to make the individual optimal to mutate with a certain probability, and to lead the group to escape from the local optimum. Convergence to global. Benchmark function test shows that the improved particle swarm optimization algorithm is superior to other optimization algorithms. The inversion accuracy of the model data and the actual seismic data is higher, the inversion section has better horizontal continuity, the vertical resolution is higher, and the interlayer deposition is better. It is proved that the algorithm has strong global searching ability and anti-noise ability. Firstly, the k-l transformation of the effect of taking off and descending dimension is used to preprocess the input attribute samples. Then the fuzzy clustering analysis provides the network parameters such as the number of hidden layer nodes, the base center and the variance. Finally, the network weight is optimized by using the improved particle swarm optimization algorithm. The practical application of standard test set data and sand body thickness prediction proves the feasibility and practical value of the network, and the prediction accuracy is high.
【学位授予单位】：东北石油大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：P631.4;TP183

【相似文献】